DenseVnet Dice loss remains and does not fall

[kye9216789]

Hi
Thank you to open your novel framework.
I used 3D DenseVnet to predict CT image label.
And before use my data, I tested your organ model and data.
But the Dice-hinge loss remains over 0.22
After I changed data to my one, also the loss remains specific value, near 0.5
I used only one data to overfit the network.
network is same with you, and I did not modified structure.

I used every parameter same with you.

I tested several time, and trained 1000 epochs but it remains as same.
Can you tell me why loss remains?

[whikwon]

Same issue occurs

[FarheenRamzan]

Same issue.

[Farihaa]

Same issue.

[KathrynSch]

Same issue

[ericspod]

This discussion on #208 may be relevant to your issues.

[anmol4210]

Hi I was also facing the same issue. But when I changed the loss function from Dice to GDSC I was able to train my network and loss also started decreasing.